Re: Statistics
Sorry, but a sample of 48 tests, roughly half-and-half control and treatment, can most certainly show statistically significant effects. Mind you, it would have been *much* better if there were at least 30 in each case, due to the central limit theorem, but having fewer largely biases the mean, not the standard deviation (until you get to very small numbers, e.g. around 4-5 in each case).
In fact, that they showed a statistically significant difference (a term of art, by the way) with a small sample, that means the effect is quite large. I'm more impressed with a study this size that shows a statiistically significant result than one with 1000 in each case. With a large sample like that, you find a statistically significant difference for *anything* pretty much. Which is why we don't use the chi-square test as definiitive with large sample sizes (in structural equation modeling, for example).
(Icon more for my field than for the mic-drop moment)
Biting the hand that feeds IT
